Validate checkpoint credentials once; log throughput

#3
by arminfg - opened
Iterate org

Uploads were retrying on every eval and 401-ing, flooding the log.

  • pass HF_TOKEN explicitly to HfApi (it was relying on ambient auth)
  • validate the token and create the repo once at startup, with a clear
    message naming the likely cause (token scoped to a user cannot write
    to an org repo)
  • disable uploads after 3 consecutive failures; training continues
  • log steps/s alongside reward
arminfg changed pull request status to merged

Sign up or log in to comment